give openbao's own start the container's long timeout

nix/host-modules/swarm-bao.nix

@@ -52,6 +52,13 @@ let
   # `nix/module-eval/grafana.nix` pins it against the shipped dashboard file.
   collectorServiceName = "bao";
 
+  # OpenBao's start job limit inside the container, and the container's own,
+  # which is the same plus a margin: the container is `Type=notify`, so its
+  # start job spans OpenBao's and must outlast it, or the container's limit
+  # kills a start the unit would still finish.
+  openbaoStartSec = 240;
+  containerStartSec = openbaoStartSec + 60;
+
   # Upstream's own default, kept so its documentation matches. The raft data
   # lives INSIDE the container on `ephemeral = false`, the same way
   # ./swarm-grafana.nix keeps its sqlite database — no sibling service binds
@@ -4138,7 +4145,7 @@ in
         # The container is `Type=notify`, so its start job waits for OpenBao,
         # whose first start after a restart takes 60-144 s opening the raft
         # storage. The 1min default kills it mid-open and every retry is cold.
-        timeoutStartSec = "5min";
+        timeoutStartSec = "${toString containerStartSec}s";
 
         inherit privateNetwork;
 
@@ -4395,31 +4402,36 @@ in
             # attrset literal and does not parse. Module merging happens across
             # `config` blocks, not within a literal — so the seal's half joins
             # with `optionalAttrs`.
-            systemd.services.openbao.serviceConfig =
+            systemd.services.openbao.serviceConfig = {
-              lib.optionalAttrs haveServerTls {
+              # Upstream sets no start limit, so the 90 s default cut off
-                LoadCredential = [ "${serverKeyCredential}:${serverKeyPath}" ];
+              # the 60-144 s raft open; `Restart=on-failure` upstream then
-              }
+              # restarts it cold.
-              # The PIN reaches openbao as an environment variable read from a
+              TimeoutStartSec = openbaoStartSec;
-              # 0400 file the provisioning unit wrote — never as a value in this
+            }
-              # expression, which would render it world-readable into the store.
+            // lib.optionalAttrs haveServerTls {
-              # `TPM2_PKCS11_STORE` is required because the store is not at the
+              LoadCredential = [ "${serverKeyCredential}:${serverKeyPath}" ];
-              # library's default location.
+            }
-              // lib.optionalAttrs (baoDeploy.seal == "pkcs11") {
+            # The PIN reaches openbao as an environment variable read from a
-                EnvironmentFile = pinEnvFile;
+            # 0400 file the provisioning unit wrote — never as a value in this
-                Environment = [ "TPM2_PKCS11_STORE=${tokenStoreDir}" ];
+            # expression, which would render it world-readable into the store.
-                # `DynamicUser` implies `ProtectSystem=strict`, which leaves the
+            # `TPM2_PKCS11_STORE` is required because the store is not at the
-                # bind mount read-only to this unit however it is owned, and the
+            # library's default location.
-                # pkcs11 library opens its sqlite store read-write.
+            // lib.optionalAttrs (baoDeploy.seal == "pkcs11") {
-                ReadWritePaths = [ tokenStoreDir ];
+              EnvironmentFile = pinEnvFile;
-                # Two groups, two different jobs: the store's sqlite file, and
+              Environment = [ "TPM2_PKCS11_STORE=${tokenStoreDir}" ];
-                # the TPM device the seal opens through it. `CapabilityBounding
+              # `DynamicUser` implies `ProtectSystem=strict`, which leaves the
-                # Set=` is empty upstream, so there is no `CAP_DAC_OVERRIDE` to
+              # bind mount read-only to this unit however it is owned, and the
-                # fall back on — group membership is the only way in.
+              # pkcs11 library opens its sqlite store read-write.
-                SupplementaryGroups = [
+              ReadWritePaths = [ tokenStoreDir ];
-                  tokenGroup
+              # Two groups, two different jobs: the store's sqlite file, and
-                  tpmGroup
+              # the TPM device the seal opens through it. `CapabilityBounding
-                ];
+              # Set=` is empty upstream, so there is no `CAP_DAC_OVERRIDE` to
-              };
+              # fall back on — group membership is the only way in.
+              SupplementaryGroups = [
+                tokenGroup
+                tpmGroup
+              ];
+            };
 
             # ⚠️ Upstream sets `restartIfChanged = false` on this unit: a restart
             # SEALS the store and disconnects every client. The container around

nix/module-eval/hive-tls.nix

@@ -853,10 +853,19 @@ let
         ) names;
     }
     {
-      # The container is `Type=notify`: its start job lasts as long as
+      # The container is `Type=notify`: its start job spans OpenBao's first
-      # OpenBao's first start, which a 1min limit cuts off mid-open.
+      # start, so OpenBao's own limit is raised too and sits a minute below
-      name = "the store's container is given five minutes to start";
+      # the container's. `Restart=` comes from upstream, so it shows the
-      ok = lib.hasInfix "TimeoutStartSec=5min" allLocal.systemd.units."container@swarm-bao.service".text;
+      # lookup reaches the real unit.
+      name = "openbao's start limit is raised and stays below its container's";
+      ok =
+        let
+          inner = allLocal.containers.swarm-bao.config.systemd.services.openbao.serviceConfig;
+          unit = allLocal.systemd.units."container@swarm-bao.service".text;
+        in
+        inner.Restart == "on-failure"
+        && inner.TimeoutStartSec == 240
+        && lib.hasInfix "TimeoutStartSec=300s" unit;
     }
     {
       name = "hive-priv is told the cache's address by the host's configuration";